Purchasing-Power-Parity Changes and the Saving Behavior of Temporary Migrants

Transcription

1 Purchasing-Power-Parity Changes and the Saving Behavior of Temporary Migrants Alpaslan Akay, Slobodan Djajić, Murat G. Kirdar y, and Alexandra Vinogradova z st November 207 Abstract This study examines how immigrants saving behavior responds to changes in the purchasing power parity between the source and host countries. For this purpose, we rst build a theoretical model of joint return migration and saving decisions of a temporary migrant, and then test the implications of this model using data from the German Socioeconomic Panel on immigrants from 99 di erent source countries. In accordance with the implications of the model, the empirical results show that immigrants saving rate decreases with a rise in the price level of the source country but increases in response to a nominal depreciation of domestic currency. In addition, the responsiveness of the saving rate is found to be higher to both variables, the shorter the period of time between the realization of the price shock and the migrant s retirement date. Abstract JEL Classi cation: Key Words: The Graduate Institute, A Avenue de la Paix, CH-2 Geneva, Switzerland. Tel: , Fax: , slobodan.djajic@graduateinstitute.ch y Bogazici University, Natuk Birkan Binasi, 2. Kat, Ekonomi Bolumu, Bebek, Istanbul, Turkey. Tel: , murat.kirdar@boun.edu.tr z Center of Economic Research, ETH, Zurich, Switzerland. Tel: , avinogradova@ethz.ch

2 Introduction An important problem facing temporary migrant workers is that of deciding how much to save while working abroad. In choosing their optimal time pro le of consumption and saving, they take into account not only the level of their wages abroad and what they expect to earn after return to the source country, but many other factors, including the relationship between foreign and home prices of consumption goods and the valuation of domestic in terms of foreign currency. The present study examines this problem using both theoretical and empirical analysis. The focus of our theoretical model is on the responsiveness of a temporary migrant s saving rate to changes in the exchange rate and the price level back home. More speci cally, we consider the impact of unanticipated changes in these price variables on a migrant s saving behavior in two distinct cases. In one case the migrant nds it optimal to return to the home country before the age of retirement and to continue working at home, while also consuming the savings accumulated abroad. We refer to this as an interior solution from the perspective of a temporary migrant s optimal timing of return. The other case is a corner solution, where the migrant returns to the home country only for the purpose of retiring and enjoying consumption at a relatively lower cost than abroad. We examine the two cases in some detail within a deterministic framework and analyze how intentions to return to the source country and the timing of the shock to the exchange rate or the price level in uence saving behavior. When an interior solution is optimal, we nd that a migrant s saving rate abroad declines with an increase in the source-country price level, but is ambiguously a ected by an increase in the exchange rate. In the special case where source-country in ation drives prices and the exchange rate up in the same proportion, the net e ect on the saving rate is negative and more so if the increase in prices is greater than the increase in the exchange rate. As we shall see in the empirical part of our paper, this in fact corresponds to the real currency appreciation experienced by the principal source countries of migration in our data set. In that case our model predicts that there should be a decline in the saving rate of migrants who 2

3 intend to return to their home country before retirement. Moreover, the magnitude of this decline should not be a ected by the amount of savings already accumulated by the migrant or by the timing of the price shock within a migrant s period of residence abroad. These results are somewhat di erent from the ones we obtain when a migrant nds it optimal to choose the corner solution for the timing of return. We nd once again that his saving rate decreases with a rise in the price level of the source country, but now his saving rate unambiguously increases in response to a nominal depreciation of domestic currency under the realistic assumption that the degree of concavity of his utility function is less than unity. Interestingly, unlike in the case of an interior solution, this increase in the saving rate is found to be larger, the shorter the period of time between the realization of the price shock and the migrant s retirement date (which in the case of the corner solution coincides with his return date). Moreover, when the price level and the exchange rate increase in the same proportion, the saving rate decreases. For a given increase in the nominal exchange rate, the decrease in the saving rate is larger if there is a real appreciation of domestic currency. We test these theoretical implications of our model using the German Socioeconomic Panel (GSOEP) for 203, which includes annual data on immigrants monthly savings in the host country from 992 to 203, as well as a rich set of information on immigrants individuallevel characteristics. We combine these information for immigrants from 99 di erent source countries with their source-country level characteristics. A peculiar feature of the GSOEP is that it also includes annual data on immigrants return intentions. This allows us to test whether the implications of our model are stronger for immigrants with stronger return intentions, as would be expected. Since the question on monthly savings is asked only for non-negative amount of savings, we use Tobit models in the estimation. In particular, we use Tobit random e ects model and the Tobit xed e ects model developed by Honore (992). The data on return intentions indicate that the majority of immigrants who intend to return in fact intend to return at or around the age of retirement. Hence, the relevant theoretical model that is relevant for testing is the model with corner solution. The empirical 3

4 evidence con rms the implications of this model. We nd that saving rate increases in the nominal exchange rate but decrease in home country prices. A 0-percent increase in the nominal exchange rate brings about a roughly -percent increase in savings. In addition, the absolute magnitudes of both relationships increases as the time that remains until retirement becomes shorter. 2 Related Literature Our study is not the rst to consider the role of price variables in in uencing the behavior of temporary migrants. Djajić (989) examines how wages and prices at home and abroad in uence a migrant s pattern of consumption and labor supply in the two economies. Those prices, however, are assumed to remain unchanged throughout a migrant s stay abroad, an assumption used in practically all subsequent theoretical contributions to the literature on the saving behavior of temporary migrants. By contrast, our focus in the present study is on the implications of unanticipated changes in the exchange rate or the price level at a point in time within a migrant s planning horizon when he is already in the foreign country and in the process of accumulating savings for the purpose of nancing consumption expenditures after return. To the best of our knowledge, the existing literature, both theoretical and empirical, has not established a causal relationship between unanticipated exchange-rate or price level shocks experienced by migrants and their saving behavior. There are, nonetheless, a number of studies that address other dimensions of migrants behavior in response to unanticipated changes in the exchange rate. Two in uential papers by Yang (2006, 2008) are prominent examples. Using the 997 Asian nancial crisis as a source of exogenous variation in the exchange rate faced by Philippino migrants in dozens of destination countries, he shows that immigrants timing of return migration, remitting behavior, and investments in the source country are See, for example, Dustmann (200), Djajić (204), Djajić and Vinogradova (205), and Vinogradova (206). See also Galor and Stark (990, 99), where Djajić (989) is extended to an analysis of the case where a temporary migrant s return date is uncertain. 4

5 signi cantly a ected by unanticipated changes in the exchange rate. 2 While migrants in Yang s studies are mostly short-term guest workers residing in dozens of host countries, our data set contains information on immigrants from numerous source countries with a wide range of residence durations in a single host country, Germany. Moreover, Yang s data is on remittance receipts and expenditure patterns of households left behind, while we observe actual earnings outcomes and saving behavior of migrants at the destination. This allows us to examine directly the impact of unanticipated exchange-rate changes on their saving rates. For migrants who intend to return to their country of origin, exchange-rate uctuations a ect the purchasing power of their foreign income as well as of their accumulated foreigncurrency savings in terms of both consumption and investment goods back home. As noted by Yang (2006, 2008) the increased valuation of foreign-currency holdings experienced by the majority of Philippino migrants during the Asian nancial crisis can potentially trigger investment in entrepreneurial activity back home as it enables the migrant household to overcome liquidity constraints they face in meeting the minimum investment requirement on a project. We focus, instead, on the impact of changes in the exchange rate and the source-country price level on a migrant s budget and his optimal time pro le of saving. In contrast with previous contributions to this literature, we develop a theoretical framework in which migrants make optimal saving and return-migration decisions in a dynamic setting. This enables us to derive theoretical predictions on how the saving rate can be a ected under various conditions by unanticipated movements in the price variables. As our data set contains information on each migrant s age, duration of stay abroad, and intentions to return to the source country, we are able to test empirically our model s predictions on how such factors interact with changes in 2 Faini (994) is an earlier study on the relationship between exchange rate shocks and remittance ows. A number of more recent empirical studies focus on various other aspects of remitting behavior of immigrants. Amuedo- Dorantes and Pozo (2006) use data on Mexican immigrants in the USA and nd that a higher income risk leads to increased remittances. Merkle and Zimmermann (992) use German data and nd that plans to return migrate are an important determinant of remittances. See also Dustmann and Mestress (200), Bauer and Sinning (20), and Sinning (20) for empirical work on the link between return intentions and remitting behavior. 5

6 the exchange or the price level in in uencing a migrant s saving behavior. Our theoretical analysis helps facilitate the choice of the most appropriate empirical speci cation, while also allowing us to interpret the estimation ndings in the context of the model s predictions. 3 As Yang does for the case of Philippino migrants, Kirdar (2009) nds that the real exchange rate a ects the return migration hazard rates of immigrants in Germany. The direction of the e ect in the two studies, however, is not the same, presumably due to the marked di erence between the two studies in terms of immigrants average duration of residence in the host country. In a follow-up paper, Kirdar (203) shows that immigrants return intentions also respond to the changes in the real exchange rate. A study by Abarcar (203) examines the relationship between exchange-rate shocks and return migration in the case of migrants residing in Australia. By contrast, our focus here is on the response of temporary migrants saving behavior to exchange rate shocks rather than the timing of return. Two more recent papers, Nekoei (203) and Nguyen and Duncan (207), investigate a causal link between migrants labor-market outcomes and real-exchange-rate shocks. As is the case with other contributions to this literature, they do not examine the implications for a migrant s saving behavior. In fact the simple income-sharing model of Nekoei (203) is based on the assumption that migrant households consume all of their current income. 4 In sum, a key distinction between the present study and these earlier contributions is that the latter lack data on migrants saving rates abroad. This prevents them from testing directly the relationship between unanticipated exchange-rate shocks and migrants saving. Instead, 3 Using a structural model of return migration and saving behavior of immigrants in Germany, Kirdar (202) also uses the variation in ppp across countries to identify the structural parameters of that model which he uses to examine the scal impact of immigrants. 4 Ngyuyen and Duncan (207) follow Nekoei (203) in examining the causal link between migrants labor-market outcomes and the exchange rate in the Australian context. While the dataset in Nekoei (203) is cross sectional, Ngyuyen and Duncan (207) exploit the panel structure of their data, which allows them to account for time-invariant unobserved heterogeneity by using xed-e ects methods. When they do not account for time-invariant unobserved heterogeneity, they nd, as Nekoei does, that immigrants reduce their labor supply in response to an appreciation of the host country s currency. Once they account for time-invariant unobserved heterogeneity, however, the negative supply response disappears. This result highlights the importance of accounting for unobserved heterogeneity. We also account for it in our empirical model and nd no evidence that exchange-rate shocks change immigrants laborsupply behavior. In our study this exercise serves merely as a check of a potential mechanism that could possibly a ect a migrant s saving rate. 6

7 they focus on establishing causal relationships between exchange-rate shocks and certain other dimensions of immigrants behavior. To the best of our knowledge, the present study is the rst to do so with regard to immigrants saving on the basis of data that are at least as rich, if not richer than those used in previous contributions. Our panel data allow us to account for heterogeneity to a higher degree, our unique data on return intentions allow us to conduct more convincing tests of the implications of our theoretical model, and the rich micro-level data allow us to conduct tests of the underlying mechanisms (through labor supply behavior) of the observed change in saving behavior. 5 3 Theoretical Framework The focus of our paper is on the e ects of unanticipated changes in the exchange rate and the price level back home on the saving behavior of temporary immigrant workers. In relation to the setting, one can think of immigrants who were recruited to meet labor shortages in Germany during its post-war economic boom. Although their migration was expected to be only temporary, many of these workers chose to stay for decades and even permanently as they were able to renew their residence permits and establish (or reunite with) families in the host country. 6 It is clear that for immigrants who intend to remain permanently in the host country, the exchange rate and the price level of the source country do not play an important role, unless they are supporting family members back home by sending remittances or plan to return periodically for the purpose of consumption on short visits. By contrast, if migration is intended to be temporary, changes in the exchange rate and the price level can have a 5 There is also a related literature on the e ect of other macroeconomic factors on the behavior of immigrants. McKenzie et al. (204) show how migrant in ows respond to GDP shocks in destination countries, Djajic et al (206) examine how emigration ows change with income in source countries, and Akay et al. (forthcoming) assess the impact on the subjective well-being of immigrants in Germany following changes in the macroeconomic conditions at the origin. 6 For immigrants recruited between 9.. and 9.. for work in Germany, X% chose to return to their countries of origin before the age of retirement, Y% returned at the age of retirement and Z% remained permanently in Germany. Murat, is there data that can be used to ll in the blanks on this footnote? 7

8 signi cant impact on a migrant s saving behavior as these price variables a ect the purchasing power of savings accumulated abroad as well as the optimal time pro le of consumption while abroad and after return to the source country. We see saving behavior of immigrants and the timing of return to the source country as elements of a solution to their problem of maximizing utility over a planning horizon. In an environment where they are subjected to unanticipated shocks, a stay abroad that is intended to be temporary may well turn out to be permanent and vice versa. In our theoretical analysis below, we refer to temporary (resp. permanent) migrants as those who intend to return to their country of origin (resp. remain in the host country). By contrast, our data set undoubtedly contains observations on migrants who intended to return, but ended up staying permanently as well as on migrants who intended to stay permanently, but nally decided to return back home. 3. A Temporary Migrant As in the case of post-war migration to Germany, let us suppose that a migrant s work/residence permit is renewable, enabling him to choose how long to remain in the host country. A migrant s planning horizon is assumed to be from the time of arrival in the host country, de ned as = 0, until = +, where is the number of years until retirement and is the duration of the retirement phase. There are two activities: (i) work and (ii) consumption of a standard basket of commodities and services. After retirement, consumption is assumed to be the only activity. While working abroad, a migrant receives at time the wage, at home he receives the home-country wage,, and he faces the price level abroad and at home when consuming goods. The exchange rate, or the price of one unit of foreign in terms of domestic currency at time, is denoted by We shall assume that the cost of consumption in the host country is higher than it is at home (i.e., ), that the foreign money wage is higher than the home wage (i.e., ), and that the real wage is higher in the host country (i.e., 8

9 ). Our migrant is assumed to be a single individual, whose problem is to maximize, the lifetime utility from consumption abroad and at home, by choosing the optimal consumption rate at each point in time from time 0 to + and the optimal return date,. Our focus is on the problem of a migrant who intends to stay temporarily in the host country. There are two possible solutions to a temporary migration problem: an interior solution, in the sense that 0 and a corner solution, =, whereby a migrant returns to the source country only for the purpose of retiring in that location. Let us begin by considering an interior solution, leaving the analysis of the corner solution for Section 3.2. To simplify the analysis and the algebra, we assume that the rate of time preference and the rates of interest at home and abroad are equal to zero. 7 Thus the objective is to maximize where c = Z 0 ( ) + Z + ( ) () and c are the time- rates of consumption abroad and after return to the source country, respectively. While abroad, the migrant saves in order to accumulate assets that later serve to support his consumption in the home country after time. Assuming that the wage rates at home and abroad are constant, the stock of assets held abroad evolves over time according to the following di erential equation: _ = where a dot over a variable indicates a time derivative. The stock of savings accumulated by the migrant in the form of foreign currency until the time of return is given by where 0 = 0 + Z 0 ( ) (2) is the initial stock of assets, net of migration costs, assumed to be held in the form of foreign currency. 8 7 The role of interest di erentials across countries and discrepancies between the rates of interest and the rate of time preference in in uencing saving decisions of temporary migrants and the optimal timing of their return to the source country is examined by Djajić (200). See also Djajić (204a, 204b), Djajić and Vinogradova (206) and Vinogradova (206). 8 The case in which savings are continuously remitted to the source country and held in the form of domestic 9

10 Let us suppose that the exchange rate and the price levels in both countries are constant over time, unless a shock occurs causing a change in one or more of these variables. The initial values of variables are denoted by the subscript 0, while the post-disturbance values have the subscript. We assume that a shock to the exchange rate or a price level is unanticipated by the migrant and that he has static expectations (i.e. any given change in the exchange rate or either of the price levels is expected to be permanent). Objective function () is maximized subject to the constraint that the value of savings accumulated abroad in the form of foreign currency until time is equal to the excess of consumption over wage earnings and retirement bene ts after return. 0 = Z ( 0 ) Z + ( 0 0 ) (3) where is the foreign-currency-denominated ow of retirement bene ts enjoyed by the migrant in the source country. Let us suppose that, as in the case of a migrant who worked in Germany, is a fraction of his foreign wage, which is increasing in the number of years spent working abroad. For simplicity, we assume that =, where is a constant. The budget constraint on the basis of which the migrant makes his decisions at = 0 concerning the optimal consumption path and the return date,, can then be written as follows: Z 0 ( 0 ) = Z ( 0 ) Z + [ 0 0 ] (4) De ning the Lagrangian associated with the migrant s maximization problem as = Z Z + ( ) + ( ) + 0 Z + ½ ( 0 ) + 0 Z ( 0 ) + Z + ¾ [ 0 0 ] currency is examined in the Appendix, where we show that the results regarding a migrant s saving behavior are qualitatively the same as under the assumption that the savings are held in the form of foreign currency. 0

11 the rst order conditions are = 0 ( ) 0 0 = 0 (5) = 0 ( ) 0 = 0 (6) = ( ) ( ) + [ 0 ( 0 ) ( 0 ) + 0 ] = 0 (7) and the budget constraint (4). These four equations enable us to solve for and the Lagrange multiplier, as functions of the the exogenous variables a ecting the migrant s behavior. Since 0 ( ) and 0 ( ) are constant in eqs. (5) and (6), the corresponding rates of consumption are also constant at 0 and 0, respectively. Having assumed that the price of the standard consumption basket is relatively higher abroad, eqs. (5) and (6) imply that as the migrant returns to the source country at =, his consumption jumps to a higher rate, while 0 ( 0 ) 0 = 0 ( 0 ) 0 0, so that the marginal utility per unit of a given currency spent on consumption is the same over the two phases of the planning horizon. To be able to derive explicit solutions in what follows, let us assume that the utility function takes the CRRA form () =, where is a measure of the degree of concavity of the utility function. In line with the available empirical evidence, our focus in what follows will be on the case of 0. 9 Using (5) and (6), we can write µ 0 = = (8) where 0 = de nes the PPP relationship at the beginning of the planning horizon. With the aid of (8), eq. (7) can be solved for 0 as a function of wages and prices that the migrant faces in the two economies and the degree of concavity of his utility function. µ 0 ( + ) 0 = ³ (9) Estimates of vary signi cantly, depending on the data used and the empirical strategy. Chetty (2006) examines some of the factors that explain this wide range of estimates. He reports that the mean estimate in the literature is = 07, while noting that studies which combine the bene ts of exogenous variation with the structural lifecycle approach, such as Blundell, Duncan, and Meghir (998), with its estimate of = 093, provide perhaps the most credible microeconomic estimates.

12 Note that when a migrant s pension is increasing in the number of years of employment in the foreign country (i.e., 0), the bene t of staying for an additional unit of time abroad also increases, as can be seen in eq. (7). This implies a higher optimal consumption rate abroad in eq. (9) and a correspondingly lower saving rate in comparison with the case where the relationship between the duration of stay abroad and the magnitude of retirement bene ts is not taken into account (see Djajić and Milbourne, 988). Also note that in the case where an interior solution is optimal (i.e., ), asset holdings do not a ect a migrant s optimal consumption rates in the two economies. As we shall see just below, initial assets in uence only the optimal duration of stay abroad. Using (8), we can also write the budget constraint (4) as ³ ( 0 0) + ( ) h i = 0 (0) which yields the solution for as a function of the consumption rate abroad and the parameters of the model, including the initial stock of assets, 0 : = ( + ) 0 0 ³ 0 () 0 ( 0 0 ) 0 0 We restrict the parameters to the range which ensures that 2 (0 ). It then simply 0 remains to introduce the optimal 0 from eq. (9) into () to determine the value of that is just su cient to enable the migrant to cover the cost of his optimal consumption program. 3.. An Unanticipated Change in PPP Our objective is to study the impact of an unanticipated change in the purchasing-powerparity relationship between the two countries on a migrant s pattern of consumption and asset accumulation. Since Yang (2006) is the rst to analyze the impact of an unanticipated exchange-rate shock on a migrant s behavior, it is useful to compare at this point the purpose of our model and that of the one presented in the Theory Appendix of Yang (2006). While we are concerned with a migrant s time pro le of consumption and saving in the host country, Yang s focus is on the implications of exchange-rate shocks for the timing of return and propensity 2

13 to invest in entrepreneurial activity at home. He does not analyze the consumption behavior of migrant workers or the implied saving behavior as his data set does not contain direct information on these variables, but rather on the ow of remittances and the expenditure pattern of the households left behind. In fact Yang assumes "that consumption overseas yields zero household utility: overseas work is a pure hardship and is done exclusively for the bene t of future raised consumption in the home country" (p.2 of the Theory Appendix). While this is a plausible assumption when modeling the behavior of Philippino guest workers on relatively short-term contracts abroad, our framework pertains to foreign workers in Germany, most of whom returned to their source country only after decades of work abroad. Moreover, as we have data on their saving rates abroad, it is important for us to consider explicitly their optimal time pro le of consumption. Another important di erence is that Yang has prices of consumption goods normalized to unity while we consider explicitly the e ects of changes in and. Moreover, in contrast with Yang (2006), the e ects of an exchange-rate shock on the optimal migration duration is not our main focus and we therefore relegate the derivation and discussion of that behavior to the appendix. In our investigation of impacts of an unanticipated change in the purchasing-power-parity relationship between the two countries, we assume that at =, while the migrant is still working abroad, there is a change in the exchange rate and/or one of the price levels that alters. We then examine how this a ects the migrant s optimal consumption pro le and the implied rate of asset accumulation. Not expecting any change in the exchange rate or price levels, the migrant follows his optimal consumption path characterized by eq. (9) and plans to return to the source country at = as given in eq. (). By the time an unanticipated change in the PPP relationship occurs at time, the migrant will have accumulated ( 0 0 ) units of foreign currency. His problem at =, when the shock to PPP is realized, is to recalculate his optimal consumption program from time to + and the optimal return date, given his asset holdings at that moment. 3

14 As can be seen in eq. (9), the stock of assets held by the migrant and the amount of time remaining within the planning horizon do not a ect the optimal consumption rate 0.0 We can then determine the impact of an unanticipated change in or on saving and consumption rates abroad by simply di erentiating (9) with respect to the relevant price variable. We also consider the implications of an unanticipated change in, as the wage in the source country may change along with the price level and the exchange rate if the economy is experiencing in ation that puts upward pressure on both prices and wages. ( 0 0 ) µ 0 0 = 0 0 ( + 0 ) 0 (2) 0 ( 0 0 ) ( 0 0 ) ( 0 0 ) 0 0 = 0? 0,? (3) 0 µ 0 =? 0, 7 (4) 0 = 0 (5) ( + ) These results concerning the migrant s consumption spending abroad imply that his saving rate declines with an increase in, but increases with an increase in in the empirically relevant range of. In addition, it is ambiguously a ected by an increase in the exchange rate and increases with an increase in. In the special case where source-country in ation drives and up in the same proportion, it can be ascertained by adding the results from eqs. (2) and (4) that the net e ect on is positive (on the saving rate negative) and even more so if the increase in is greater than a given increase in. As we shall see in the empirical part of the paper, this in fact corresponds to the behavior of the exchange rate and the price level in the principal source countries of migration in our data set. We should therefore expect that in such cases of real appreciation of source-country currency the saving rate of migrants who intend to return to their home country before retirement will tend to 0 Note that our focus is on an environment in which the migrant chooses an interior solution for. In that case initial asset holdings a ect the optimal return date, but not the optimal rates of consumption, which are determined by conditions (5)-(7). By contrast, asset holdings will clearly have an e ect on when we consider parameters of the model for which the migrant chooses to return to the source country for the purpose of retirement (i.e., = ). We examine that case in the next section. 4

15 decline. Note, in addition, that if the increase in and is in the same proportion, leaving the PPP relationship and the real wage at home una ected, this has no impact on a migrant s saving rate (i.e., the sum of expressions in eqs. (2), (4), and (5) is zero). 3.2 Return for Retirement Only Conditions in the labor and goods markets at home and abroad may be such that it does not pay to return to the source country before time. This can well be the case if a worker migrates late in the planning horizon (small ) or if the international wage di erential in favor of the host country is large, while the price-level di erential o ers a considerable advantage to a migrant who consumes at home rather than abroad over the retirement phase of the planning horizon. More speci cally, a temporary migrant chooses the corner solution when the value of that satis es condition (9) and the corresponding rate of consumption after return to the source country (as given by condition (8)) are not attainable within the migrant s budget even if he spends the rest of his working life abroad. Then he must choose a lower time pro le of consumption, as dictated by conditions (5) and (6) and the budget constraint (4) (with the duration of stay abroad set at = ). Our data set contains information on the intentions to return, revealing that x% of migrants in our sample indicate at least once that they intend to return to their country of origin, while y% indicate more then 50% of the time that they intend to return. The advanced age at the point of return for the majority of migrants (see Table...), suggests that the planned return was simply for the purpose of retiring in the source country. (Murat, can you ll in the values for X and Y at the beginning of the paragraph and perhaps introduce a table referred to above if you think it is appropriate.) In that case the migrant s optimization program, with the return date at =, is as follows: = Z 0 ( ) + Z + ( ) (6) 5

16 subject to the budget constraint Z 0 ( 0 ) = Z + ( 0 ) (7) where is the fraction of the foreign wage that the migrant expects to receive in the form of pension bene ts after having worked abroad for years. The solution to this problem yields the constant optimal consumption rate abroad prior to any shock to the PPP relationship between the two countries: 0 = ( + ) (8) 0 The solution for the constant consumption rate at home over the retirement phase of the planning horizon is, as in the previous section, 0 = If there is an unanticipated change in PPP at =, a migrant will adjust his optimal consumption rates at home and abroad in response to this change in the environment. Denoting once again the pre-disturbance values of variables by the subscript 0 and the postdisturbance values by the subscript, a migrant s optimal consumption rate after return to the home country is =, while the optimal consumption rate abroad is the solution for that satis es the following budget constraint. We thus have [ 0 + ( 0 0 )] + ( ) ( ) + [ ] = 0 (9) To examine the sensitivity of = 0 + ( 0 0 ) + [ + ] ³ (20) + to unanticipated changes in the exchange rate and the price levels at time, we di erentiate eq. (20) with respect to and : 6

17 ( ) ( ) ( ) = + = + = +? 0,? (2)? 0,? (22)? 0, 7 (23) where refers to the PPP relationship following the shock to the corresponding variables. With the empirically relevant value of being less than unity, these expressions indicate that, a migrant s nominal rate of consumption spending abroad, decreases (saving rate increases) if the home currency depreciates or the foreign price level rises and increases (saving rate decreases) with an increase in the price level of the source country. Proposition : Suppose that A migrant s saving rate abroad (i) increases in response to home-currency depreciation and to an increase in the foreign price level;(ii) decreases in response to an increase in the domestic price level. When and rise in the same proportion, the e ect on becomes: ³ = + + ( ) + ( ) + + = + + indicating that consumption abroad increases (saving rate decreases) in the empirically relevant case where. Moreover, if, as in the majority of source countries in our sample over the time period under consideration, our analysis suggests that the increase in the consumption rate abroad (decrease in the saving rate) should be even larger for any given nominal rate of currency depreciation. These results are qualitatively similar to those we present in the previous section concerning the response of a migrant who chooses an interior solution. Note, in addition, that movements in the source-country wage have no impact on as a migrant has no intention of participating in the labor market of the source country if he chooses the corner solution. 7? 0, 7

18 As may be seen in eqs. (2), (22), and (23), the impact on of any given unanticipated change in or depends on the point in time along a migrant s planning horizon at which the unanticipated shock occurs. This is in contrast with our ndings in the previous subsection, where the change in is found to be independent of the timing of the unanticipated shock to PPP. In what follows, we refer to as the number of years since migration (YSM for short). The e ect of YSM on the relationship between consumption and PPP is of particular interest if we seek to understand di erences in the saving behavior among various cohorts of immigrants. To examine this relationship, we di erentiate eqs. (2), (22), and (23) with respect to, which yields: µ ( ) µ ( ) µ ( ) = = = h h h i 2? 0,? (24) i 2? 0,? (25) i 2? 0, 7 (26) The condition is both necessary and su cient for (24) and (25) to be negative. In that case, the decrease in the consumption spending abroad (and hence the increase in the saving rate) in response to an unanticipated increase in the exchange rate or the foreign price level is larger, the greater the value of relative to, where is the number of years from the time of migration to retirement. Thus the shorter the period of time between the realization of the PPP shock and the migrant s retirement date, the greater the proportional change in the migrant s consumption rate abroad and the corresponding change in his saving rate. To see the intuition behind this result, let us turn to eq. (24) which relates to the interaction between the e ect on of a change in the exchange rate and. Note that when, re ecting a relatively high degree of substitutability between consumption abroad and consumption at home, the increase in nominal spending at home is proportionately greater than the increase in, for any given, as indicated by eq. (8). This implies that more foreign currency is 8

19 needed to cover the optimal rate of consumption over the years of retirement after return. To support that higher optimal rate of spending, the saving rate abroad has to increase and increase more, the shorter the remaining period of stay abroad before retirement (i.e., the greater is for a given T). In sum, for the empirically relevant case of, the reduction in the migrant s foreign consumption rate is larger, the closer is the date of the shock to the retirement (and hence return) date. Accordingly, as a result of an unanticipated increase in the exchange rate, we should expect to see a larger increase in the saving rate of those migrants who have been abroad for a relatively longer period of time, other things being equal, including a worker s age at the time of migration. The same line of reasoning can be invoked to explain eqs. (25) and (26), which state that the response of the migrant s consumption rate (saving rate) abroad to a change in the foreign or the home price level is stronger (weaker) the larger is relative to. We summarize the results in Proposition 2: Suppose that The response of the migrant s saving rate to changes in the exchange rate or the price levels at home and abroad is stronger as the number of years until retirement and return migration becomes smaller. These ndings are in sharp contrast with the presumption that an appreciation of foreign currency makes a migrant "wealthier" in the sense of increasing the purchasing power of the savings accumulated in the form of foreign currency, so that he can reduce his saving rate for the remainder of his stay abroad and still meet his expenditures during the retirement phase at home. Reasoning along these lines ignores the fact that an increase in also creates a larger wedge between the optimal values of and, which entails an increase in the foreigncurrency value of the savings needed to support the optimal consumption rate for the years of retirement after return. Hence the shorter the time period over which these additional savings can possibly be accumulated abroad, the larger must be the drop in. Let us nally consider an environment in which in ation in the source country drives both the exchange rate and the price level higher at the same rate. For the case of, we nd 9

20 that a migrant s saving rate abroad increases, the smaller the di erence between and. ( ) + ( ) ( ) = ³ + 2? 0, 7 (27) The intuition behind this result is along the same lines as outlined in the previous paragraph. 4 The Evidence 4. Data The micro-level data in our empirical analysis come from the German Socio-Economic Panel (GSOEP). It is a large and nationally representative panel data of households in Germany; around 30,000 individuals in,000 households are surveyed each wave. The data include foreigners and recent immigrants to Germany, as well as Germans. In fact, the initial wave in 984 started with an oversample of foreigners in Germany from ve main source countries (Turkey, ex-yugoslavia, Greece, Italy, and Spain). Immigrants from these countries still form the bulk of the immigrant sample in GSOEP. We use the 203 version of GSOEP, which includes annual data from 984 to 203. An advantage of the GSOEP is that it is very rich with regards to socio-demographic and economic characteristics of individuals. Since our dependent variable, monthly savings, is available at the household level, we need to conduct our analysis also at the household level. Hence, we extract all immigrant households in all subsamples of the GSOEP. We de ne immigrant households as those whose head is an immigrant. In turn, an immigrant household head is de ned as an individual who arrived in Germany after age 8. (Since we interpret return migration as part of optimal life-cycle decisions, the individual must have made the decision to migrate himself/herself.) We exclude household heads who arrived in Germany after age 8 but who are Germans who previously lived abroad. (This information for this tiny fraction of individuals is available in the immigration biography of individuals.) Our nal sample includes 4,498 households with an immigrant head. 20

21 We put the data into person-year format for these immigrant household heads. We follow these from the time they enter the data to the time they drop from the sample or until 203. We drop person-year observations in which the household head is aged 65 or above (in accordance with the retirement age in Germany). In addition, we drop immigrants from countries where purchasing power parity averages below in the data (all of which are developed countries) because the purpose of the immigration of these individuals to Germany could not be accumulating savings. The remaining sample includes immigrants from 2 countries. The key piece of information that comes from GSOEP is immigrants monthly savings. Immigrants are asked about how much they save on average monthly for larger purchases, emergency expenses or to acquire wealth. 2 However, this variable in censored below at zero because households are not asked about dissaving. This question was introduced to the survey in 992 for the rst time. Hence, we have data on monthly savings for the period. The other variables that come from GSOEP include years since migration, household income, household size, and dummies for the following outcomes: unemployed, married, spouse abroad, child abroad. All prices (monthly savings, household income) are normalized in 200 Euros. GSOEP also includes a peculiar question on immigrants willingness to return to their home countries. If an immigrant indicates an intention to return, he/she is also asked about the intended duration of residence in Germany in years. We also utilize this information in our empirical analysis in distinguishing between immigrants who intent to return and who do not. Moreover, by generating the intended age of return, we distinguish between immigrants who intend to return before retirement and those who intend to return after retirement. We combine our micro-level dataset with a number of auxiliary datasets. Data on purchasing power parity and exchange rates of source countries with respect to Germany come from the World Development Indicators (WDI) database of the World Bank. Data on the consumer price index in Germany also come from the WDI. We combine these three pieces of These countries are Norway, Denmark, Japan, Switzerland, New Zealand, Sweden, Australia, Finland, Ireland, Great Britain, Luxembourg, France, and Holland. 2 The exact wording of the question is as follows: "Do you usually have an amount of money left over at the end of the month that you can save for larger purchases, emergency expenses or to acquire wealth? If yes, how much?" 2

22 information to calculate the consumer price index in each source country. The nal piece of data from the WDI is GDP per capita (in constant 200 US dollars) for all source countries in the sample. Finally, we obtain data on political violence at the country level from the MEPV dataset. This dataset includes information on both interstate con ict and societal con ict. Interstate con ict covers international violence and international warfare, whereas societal con ict covers civil violence, civil warfare, ethnic violence and ethnic warfare. Each item is given a score from (lowest) to 0 (highest). We use the aggregate political violence score, which is the sum of these six items. 4.. Descriptive Statistics Table provides descriptive statistics on individual-level characteristics in panel (A) and on country-level characteristics in panel (B). Individual-level characteristics are further divided into two panels; panel (A) gives descriptives for the 3,084 individuals in the sample whereas panel (A2) gives descriptives for the,643 person-age observations across the panel. According to panel (A), the mean age at arrival is 30 and 65 percent of the household heads are male. Panel (A2) shows that 44 percent of the immigrants have positive savings and the mean amount of monthly non-negative savings is about 270 Euros. Given that monthly income is about 2,500 Euros, the saving rate (with the censored saving variable) is above 0 percent. In the panel, the average years since migration is 8 years and the average age is 47. While the fraction of observations in which individuals are married is 79 percent, the majority of the spouses and underage children reside in Germany. In terms of country-level characteristics, panel (B) shows that the average purchasing power parity is Figure A: PPP of Selected Countries with Germany Figure A2: Exchange Rate of Selected Countries with Germany Figure A3: Log Price Level in Selected Source Countries Figure A4: Log Price Level in Germany 22

23 4.2 Empirical Speci cation and Estimation In order to test for the implications of the theoretical model regarding the relationship between immigrants savings with the exchange rate, the price level in the home country and the price level in Germany, we use the following empirical speci cation, = (28) where is monthly savings of individual i at time t, is the exchange rate between Germany and individual i s home country at time t, is the price level at time t in the home country of individual i, is the price level in Germany at time t, stands for the set of control variables for individual i at time t, stands for time dummies, and is the error term. According to our model, we expect our key parameters of interest and 3 to be positive and 2 to be negative. To test the implications of our model regarding how the a ects of our key macro-level variables change by years since migration, we modify the above speci cation as follows, = ( ) ( ) (29) + 5 ( ) where is years since migration for individual i at time t and is the error terms. Here, in accordance with our model, we expect 2 and 5 to be positive and 3 4 to be negative. The control variables,, include the key characteristics of the household and household head pertaining their saving behavior: household income and household size (both in logarithmic form), dummies for unemployment, married, child abroad, and spouse abroad status of the household head, as well as duration of residence of the household head in Germany in quadratic form. We would expect savings to increase in household income but decrease in household size, and to decrease in unemployment status of the household head but to increase when the household head has a child or spouse living in the home country (which would stand 23

24 for a higher likelihood of return migration and, therefore, imply higher savings in the host country). A potential speci cation concern in equation (28) is that our key macro-level variables could partly stand for other macro-level variables that also have a bearing on the saving rate. For instance, if there is a economic crisis in Turkey, not only the exchange rate and prices in Turkey would change but also family members back in Turkey could demand more remittances due to their lower income which would need to come from migrants savings. Similarly, an unexpected political con ict in Turkey could not only in uence economic conditions and therefore the exchange rate with Germany, but also immigrants return propensity to Turkey and, therefore, their savings. Hence, the control variables in X also include the GDP per capita in the source countries (in logarithmic form) as well as a political con ict index. Macro-levels shocks in Germany could also be confounding the e ects of our key macro variables in equation (28). Suppose that a negative economic shock changes natives perception of immigrants in Germany. In that case, immigrants propensity to return to their home country and, therefore, their saving behavior would change. This negative economic shock would also in uence the exchange rate and prices in Germany. To account for these kind of shocks, we include calendar year dummies. These dummies which are common for immigrants from di erent countries capture the e ect of macro-level shocks in Germany. Finally, when we interpret equation (28) as a di erence-in-di erences framework where we compare countries over time, we are making the common-trend assumption across countries in savings. However, if there are di erent trends in savings across countries and the degree of trend is correlated with the change in macro-level variables, we would have a speci cation problem. To account for this possibility, we also add country-speci c time trends to equation (28) as a robustness check. We estimate equations (28) and (29) using four di erent panel data estimation methods. In the rst two, we ignore the censored nature of savings, that is we consider savings which are negative and zero as a part of the continuous savings distribution, and carry out xed-e ects 24